A mathematical approach to radiographicimage diagnosis -cystic lesions-

要对口腔病变作出准确的诊断，必须综合分析临床症状、体征、既往病史以及各种检查结果，包括放射学检查和组织病理学检查。医生需要多年的临床经验，才能从大量的信息中进行选择，并只使用有用的信息。然而，如果积累的信息能够以更有用和更容易获得的形式分类，并能够快速参考相关病例，这将在一定程度上缩小医生之间的经验差异，提高诊断的准确性。我们输入患者的年龄、性别、病变部位、疾病和图像数据作为基本数据，并计算在基于这些数据的计算机辅助医疗决策中正确诊断的概率。15,235名…患者中13,036名的数据以1987~1992年在大阪牙科大学口腔放射科就诊的患者为研究对象，建立炎症、牙源性、非牙源性、牙源性肿瘤、非牙源性肿瘤、恶性肿瘤、创伤、颞下颌关节紊乱病、无法分类疾病等10大类68个病种。病变部位分为7类32个部位。X线影像分为6类(18种类型)。X线表现以颌骨透亮最多见，占86%，炎症占86%。从这些数据中，我们建立了骨内病变的症状-疾病矩阵，并计算了后验概率，统计决策试图从这些数据中对289例组织病理学类型造釉细胞瘤的X线表现进行后验概率的计算，后验概率的计算结果基本正确。但是，这一不同寻常的放射检查发现案例并不正确。这些后验概率被认为有助于医学决策的放射学图像的解释。较少

Accurate diagnosis of oral lesions can be made only by corn-prehensive evaluation of a large quantity of information including in clinical signs and symptoms, and the past history as well as findings on various examinations including radiographical and histopathological examinations. Clinical experience over many years is needed for a physician to become able to select from a large volume of information and utilize only that which is useful However, if the accumulated information can be classified in a more utilizable and accessible forms and can be quickly referred to concerning the case in question, it would, to some extent, diminish the difference in experience among individual doctors and improve the accuracy of the diagnosis.We input the age, sex, site of the lesion, disease, and image data of patients as basic data, and calculated the probability of the correct diagnosis in computer-aided medical decision making based on these data. Data of 13,036 of the 15,235 patients who were … More examined at the Department of Oral Radiology, Osaka Dental University during the 8 years between 1987 and 1992 was classified and used as basic data.Ten disease categories (68 lesions), i.e., inflammation, odontogenic cysts, non-odontogenic cysts, odontogenic tumors, non-odontogenic tumors, malignant tumors, trauma, TMJ disorders, unclassifiable diseases, and others were established. The sites of the lesions were classified into 7 categories (32 sites). Radiographic images were classified into 6 types (18 patterns). Radiographic images were most frequently radiolucencies of the jaws, which were observed in 64% of the patients, and inflammatlon accounted for 86%. No changes, which were the next most frequent radiographic images, were observed in the jaw bones in 14%.We derived a symptom-disease matrix of intra-osseous lesions from these data and calculated the posterior probability and the statical decision tried to radiographic findings from histopathological type of ameloblastoma (289 cases) from these data and calculated the posterior probability.The results of posterior probability were almost correct. But, the unusual radiographic finding case was not correct. These posterior probability are considered to assist interpretation of radiographic images for medical decision making. Less